search

cvhciKIT / sloth

Sloth is a tool for labeling image and video data for computer vision research.
Other
608 stars 198 forks source link

where to get okapy #89

Closed shiquanwang closed 7 years ago

shiquanwang commented 9 years ago

I searched around but can't find okapy for install (pypi, in this project, google, etc..)

Please help

baeuml commented 9 years ago

okapy is an internal library used in our group. Theoretically you shouldn't need it and sloth should fall back to using PIL for loading images if it can't find it.

If you are using videos, you can try #87 .

I agree that the dependency on okapy is less than optimal, especially since we can't open source it for various reasons. I'd like to remove the okapy dependency completely at some point, probably best by abstracting out video and image sources, but haven't gotten around to do it yet... :(

herr-biber commented 9 years ago

The API of the okapy stuff should be self-explaining. So feel free to replace those calls with other established python libraries.

baeuml commented 9 years ago

@shiquanwang is it just an installation problem or does the loading of images fail even when you have PIL installed ?

shiquanwang commented 9 years ago

@baeuml the installation is smooth. I need to annotate a video file and okapy is needed in the error message when add the file.

baeuml commented 9 years ago

OK. Did you manage to try PR #87? Does it work for you?

shiquanwang commented 9 years ago

@baeuml Yes I tried and not working.

baeuml commented 9 years ago

:( Could you maybe post the error/problem in the thread of #87 ? Maybe Alex can take a look.

Another option would be for you to burst your video into single frames and work those images. That is actually what I did most of the time because one has to deal with video decoding only once, while bursting.

You can then add all single frames via the command line to your label file:

 sloth appendfiles [options] <labelfile> <file1> [<file2> ...]

for example:

 sloth appendfiles mylabels.json *.jpg

HTH until we have a working video solution.